Analysis of critical factors to asphalt overlay performance using gradient boosted models

Abstract

Traditional pavement performance predictions based on empirical equations typically utilize only a limited number of parameters to form relatively simple models, thus fail to generate predictions of satisfactory accuracy. With the ever-increasing size of pavement data, model-free methods such as analytical learning algorithms have become promising alternatives. Therefore, to improve the accuracy of pavement performance prediction and take advantage of the existing large dataset, this study introduced an analytical approach, the gradient tree boosting model (GTBM), to predict asphalt overlay performance and identify key factors affecting it. Five indicators were selected to represent the overlay performance, including roughness (in international roughness index, IRI), rutting, fatigue cracking, transverse cracking, and longitudinal cracking. All data were collected from the Specific Pavement Studies 5 (SPS-5) in the Long-Term Pavement Performance (LTPP) program. The results showed that pre-overlay rutting and transverse cracking were crucial to the development of overlay performance, thus pretreatment of repairing existing rutting and transverse cracking on the old pavement is essential to prolong the service life of asphalt overlay. The initial value of overlay IRI was the most critical factor for the IRI prediction. Besides, overlay roughness was found to be strongly related to the asphalt content and gradation of hot mix asphalt (HMA); fatigue cracking exhibited a close relationship with the thickness of underlying pavements and asphalt viscosity; transverse cracking was strongly associated with the air voids of HMA; non-wheel path longitudinal cracking was relevant to the asphalt content and gradation of HMA.